Under the hood of the AI economy: Databricks CEO Ali Ghodsi

Speakers

Deirdre Bosa — Anchor, CNBC TechCheck

Ali Ghodsi — CEO, Databricks

Summary

Databricks recently raised $7 billion despite being free cash flow positive, specifically to build a "war chest" against a potential tech market crash similar to 2000 or 2022.
A major shift is occurring from chatbots to "AI Agents" that build software; 80% of databases on Databricks are now created by AI, not humans.
The "System of Record" SaaS business model is under existential threat as AI agents replace the two traditional moats: user interface (UI) stickiness and database lock-in.
Chinese open-source models (like DeepSeek and Kimi) are creating intense deflationary pressure on US model pricing, forcing margins down across the sector.

Trade Ideas

Ticker	Direction	Speaker	Thesis	Time
OPEN SOURCE MODELS /CHINESE AI	WATCH	Ali Ghodsi CEO, Databricks	Chinese models and open-source alternatives are catching up to US closed models rapidly. Models like "Kimi" and "DeepSeek" are performing nearly as well as top-tier US models but at a fraction of the cost (or free). This creates a "race to the bottom" for pricing power among US model providers. Databricks' largest customers are offloading workloads to cheaper Chinese/open models for cost efficiency. Geopolitical regulation or chip bans could stifle the progress of Chinese models.	—
SYSTEM OF RECORD /LEGACY SAAS	AVOID	Ali Ghodsi CEO, Databricks	Traditional "System of Record" software companies (SaaS) are facing a "wipeout" scenario similar to the dot-com bust if they do not adapt immediately. These companies historically relied on two moats: a) The Interface Moat: Humans were trained on complex UIs, making switching costs high. AI Agents now use natural language, rendering the UI irrelevant. b) The Database Moat: Moving data was hard. New "Lakehouse" architectures allow AI agents to query data anywhere, breaking vendor lock-in. If a company charges based on "seats" (human users), their revenue will collapse as one AI agent replaces 10,000 human users. Databricks sees 80% of new databases being built by AI agents. Investors are privately questioning the efficiency and survival of traditional SaaS metrics behind closed doors. Incumbents with massive distribution might successfully pivot by integrating AI fast enough to protect their revenue base.	1:46
DATA CENTERS /AI HARDWARE	WATCH	Ali Ghodsi CEO, Databricks	There is unprecedented capital expenditure ($50B–$100B) flowing into hardware, data centers, and energy. While the AI trend is real, the current build-out creates a risk of "overbuilding." The market is pricing in perfection, but physical constraints (energy) and ROI questions remain. The sheer volume of capital chasing hardware, combined with "circular" funding deals in the AI startup ecosystem, mirrors the pre-crash vibes of 2000. If AI adoption accelerates faster than hardware supply, these stocks will continue to run despite valuation concerns.	11:51
PROFESSIONAL SERVICES /CONSULTING	SHORT	Ali Ghodsi CEO, Databricks	Enterprise clients are using AI to aggressively squeeze costs out of vendors, auditors, and consultants. Tasks that used to justify high fees (e.g., analyzing earnings calls, auditing financial data) can now be done by AI agents in minutes. Clients are demanding fee reductions because they know the vendor's cost of labor has dropped, or they are bringing the work in-house. KPMG was pressured by clients to lower fees because AI made their auditing work cheaper. RBC analysts now use AI to synthesize earnings calls in 15 minutes, work that previously took days. High-end strategic consulting may remain insulated if it relies on human relationships and complex judgment rather than data processing.	—